Thread by @WittyUsername30

Some highlights / takeaways / thoughts / comments from #RWRI 14, Day 2:

1. There are two major types of distributions: Mediocristan (e.g. height) and Extremistan (e.g. wealth).

In Mediocristan, the average of a small sample (n=100 or even n=10) will be reasonably close to the average of the entire population. If you take the average height of 100 people, you won't be too far off the average height of everyone in the world.

In Extremistan, the average of almost any sample is worthless, because a few data points contain almost all of the data. The average wealth of 100 people is nowhere near the average wealth of everyone in the world, since Gates / Bezos / etc. hold such a huge portion of wealth.

In short, you can't be 10x taller than someone else, but you can be 1,000,000x wealthier (or more).

Computing the mean, variance, and regression is useful in Mediocristan, because you can get a representative sample. The problem is that much of life is lived in Extremistan, where mean, variance, and regression are meaningless.

So a "study" that maps IQ (Mediocristan) to wealth (Extremistan) doesn't work. Yet comparisons like this are made all the time.

This is why much of behavioral economics is worthless: they are zoomed in on a single decision (Mediocristan), but life is lived (and decisions are made) within the context Extremistan.

You can't use tools for Mediocristan to analyze Extremistan.

2. Similar to a takeaway from Day 1, summary statistics such as mean and variance are nothing compared to a deep understanding of the distribution / situation.

3. People go bust all the time. Much of the time, it's not because they have the wrong idea, but because they have the wrong strategy.

"You can go bust with a favorable bet."

Even if you have a stock picking model with a huge edge, you'll go bust unless you understand how to size your bets.

In a sense, finding an edge (in whatever you do) is the easy part. Surviving the variance long enough to see your edge play out is the hard part.

4. Most people (esp economists) view the 'worst case scenario' as death of the individual. But there are outcomes that are worse than your own death:

- Your own death + death of your loved ones
- Your own death + death of humanity

This is why comparing car accident deaths with pandemic deaths is moronic. Car accident deaths can't kill the entire population; pandemics can.

5. Another distinction between Mediocristan and Extremistan is ergodicity.

Something is ergodic if you can derive an expectation for the population by observing one instance over time.

For example, if you observe one coin being flipped 100 times, you'll have a realistic expectation for the result of 100 coins being flipped once. This is ergodicity.

If you observe one (non-card counting) person playing blackjack for 100 days, you'll see them go bust (-100% return). But the total return of 100 people playing blackjack for one day will be much better than -100%. This is non-ergodicity.

The driving difference between ergodic and non-ergodic is the presence of an absorbing barrier. When you play blackjack, you can go to zero (absorbing barrier), which ends the game before you get to Day 100. Whereas when flipping a coin, there is no result that will end the game.